Add .NET BYOK E2E coverage

[roji]

The SDK's E2E suites currently validate only the CAPI path, leaving BYOK protocol regressions undetected until later integration stages. This adds .NET coverage for Anthropic Messages, OpenAI Responses, and OpenAI Chat Completions while continuing to use the existing conceptual snapshot corpus. We do this for .NET since we run that in the runtime repo as well for early regression discovery.

• Adds three E2E test legs to the .NET workflow; each leg takes about 4-5 minutes total and runs in parallel.
• Clearly CAPI-specific tests have a CapiOnly trait for filtering them out. Failing tests that look like they should be passing on BYOK currently have TODO(BYOK) comments for later investigation.
• The same snapshots used for CAPI are used for BYOK too, with the replay proxy modified to support other protocols etc.

---

Approach

• Add protocol adapters that normalize BYOK requests into the existing canonical Chat Completions representation and render protocol-complete JSON/SSE responses back to the runtime.
• Select one protocol descriptor per backend so CAPI and BYOK share endpoint routing, canonical matching, errors, response delivery, and exchange inspection.
• Keep snapshots model-agnostic and reusable across all backends. CAPI remains the sole snapshot-capture path; BYOK replay is explicitly read-only, including compaction replay through the normal snapshot matcher.
• Select the E2E backend centrally in the .NET fixture, preserving providers and models explicitly configured by individual tests.
• Add three Ubuntu in-process BYOK legs alongside the existing CAPI matrix without multiplying them across transports or operating systems.
• Classify CAPI-only and self-configured-backend tests explicitly, and add path-scoped authoring guidance so future .NET E2E tests continue to exercise every provider.

The shared replay harness remains protocol-neutral despite retaining its existing TypeScript class name for compatibility with other SDKs. The .NET wrapper is renamed to ReplayProxy.

Validation

• Replay harness: 54 tests passed; TypeScript type-check passed.
• Full CAPI .NET baseline after replay unification: 681 passed, 3 skipped.
• Anthropic Messages compatible .NET E2E suite: 458 passed, 3 skipped.
• Streaming and compaction replay scenarios passed for CAPI and all three BYOK protocols.
• Existing CAPI baseline and all three compatible BYOK suites passed before the final harness simplification; the remaining local full Responses failures reflect current CLI tool-catalog drift against older snapshots (apply_patch/rg versus edit/grep), not response-envelope rendering.

[Copilot]

Pull request overview

Adds .NET BYOK E2E coverage by adapting Anthropic and OpenAI protocols to the canonical replay snapshot format.

Changes:

• Adds Anthropic Messages and OpenAI Responses adapters.
• Centralizes .NET E2E backend/provider selection.
• Adds three Ubuntu in-process BYOK CI legs and backend traits.

Show a summary per file

File	Description
test/harness/responsesApiAdapter.ts	Adapts OpenAI Responses requests and responses.
test/harness/replayingCapiProxy.ts	Adds protocol-aware, read-only BYOK replay.
test/harness/modelProtocolAdapterShared.ts	Defines shared adapter utilities.
test/harness/modelProtocolAdapters.test.ts	Tests adapters and backend replay.
test/harness/anthropicMessagesAdapter.ts	Adapts Anthropic Messages requests and responses.
dotnet/test/Unit/E2ETestBackendTests.cs	Tests backend configuration.
dotnet/test/Harness/ReplayProxy.cs	Renames and configures the replay proxy wrapper.
dotnet/test/Harness/E2ETestContext.cs	Applies selected backends to sessions.
dotnet/test/Harness/E2ETestBase.cs	Routes session operations through context.
dotnet/test/Harness/E2ETestBackend.cs	Defines backend configuration and traits.
dotnet/test/E2E/ToolsE2ETests.cs	Uses backend-aware session creation.
dotnet/test/E2E/TelemetryExportE2ETests.cs	Uses backend-aware session creation.
dotnet/test/E2E/SuspendE2ETests.cs	Uses backend-aware create/resume helpers.
dotnet/test/E2E/SubagentHooksE2ETests.cs	Uses backend-aware session creation.
dotnet/test/E2E/StreamingFidelityE2ETests.cs	Adds backend classification and resume routing.
dotnet/test/E2E/SessionFsSqliteE2ETests.cs	Routes sessions through context.
dotnet/test/E2E/SessionFsE2ETests.cs	Routes create/resume operations through context.
dotnet/test/E2E/SessionE2ETests.cs	Adds backend classifications and resume routing.
dotnet/test/E2E/SessionConfigE2ETests.cs	Classifies provider-specific scenarios.
dotnet/test/E2E/RpcTasksAndHandlersE2ETests.cs	Marks a CAPI-only scenario.
dotnet/test/E2E/RpcSessionStateExtrasE2ETests.cs	Classifies model/provider tests.
dotnet/test/E2E/RpcSessionStateE2ETests.cs	Uses backend-aware session creation.
dotnet/test/E2E/RpcServerE2ETests.cs	Classifies and routes RPC session tests.
dotnet/test/E2E/RpcMcpAndSkillsE2ETests.cs	Routes sessions through context.
dotnet/test/E2E/RpcExtensionsLoadedE2ETests.cs	Routes extension sessions through context.
dotnet/test/E2E/ProviderEndpointE2ETests.cs	Classifies provider endpoint tests.
dotnet/test/E2E/PerSessionAuthE2ETests.cs	Marks authentication tests CAPI-only.
dotnet/test/E2E/PermissionE2ETests.cs	Uses backend-aware resume handling.
dotnet/test/E2E/PendingWorkResumeE2ETests.cs	Routes pending-work sessions through context.
dotnet/test/E2E/MultiProviderRegistryE2ETests.cs	Marks self-configured provider coverage.
dotnet/test/E2E/MultiClientE2ETests.cs	Routes multi-client sessions through context.
dotnet/test/E2E/MultiClientCommandsElicitationE2ETests.cs	Routes multi-client command sessions.
dotnet/test/E2E/ModeHandlersE2ETests.cs	Marks mode-handler tests CAPI-only.
dotnet/test/E2E/ModeEmptyE2ETests.cs	Uses backend-aware session creation.
dotnet/test/E2E/GitHubTelemetryForwardingE2ETests.cs	Marks telemetry forwarding CAPI-only.
dotnet/test/E2E/CopilotRequestWebSocketE2ETests.cs	Marks self-configured WebSocket coverage.
dotnet/test/E2E/CopilotRequestSessionIdE2ETests.cs	Classifies CAPI and BYOK session-ID tests.
dotnet/test/E2E/CopilotRequestCancelErrorE2ETests.cs	Routes request-handler sessions through context.
dotnet/test/E2E/ClientOptionsE2ETests.cs	Applies backend-aware session helpers.
dotnet/test/E2E/ClientE2ETests.cs	Integrates the E2E context fixture.
dotnet/test/E2E/ByokBearerTokenProviderE2ETests.cs	Marks explicit BYOK provider tests.
dotnet/test/AssemblyInfo.cs	Updates replay proxy terminology.
.github/workflows/dotnet-sdk-tests.yml	Adds BYOK CI matrix legs and filtering.
.github/instructions/dotnet-e2e.instructions.md	Documents backend-aware E2E authoring.

Review details

• Files reviewed: 44/44 changed files
• Comments generated: 3
• Review effort level: Medium

[github-actions]

SDK Consistency Review

Summary: This PR adds BYOK (Anthropic Messages, OpenAI Responses, OpenAI Chat Completions) E2E test coverage to the .NET SDK harness. It does not modify any SDK public API surface.

Scope

All 44 changed files are in:

• .github/ — CI workflow and instructions
• dotnet/test/ — .NET test harness and E2E tests
• test/harness/ — Shared TypeScript replay harness (new protocol adapters)

No SDK source code is modified in any language.

Cross-SDK Consistency Assessment

✅ No API consistency issues. The public API surface across all six SDKs (Node.js, Python, Go, .NET, Java, Rust) is unchanged.

Informational: BYOK E2E Coverage Gap

The shared replay harness (test/harness/replayingCapiProxy.ts) already supported all four backends (capi, anthropic-messages, openai-responses, openai-completions) via ReplayBackend prior to this PR — the new adapter files just fill in the implementation. This PR wires up .NET to exercise those backends in CI.

The other SDKs (Node.js, Go, Python, Java, Rust) do not yet have equivalent BYOK backend E2E test matrix legs. This is not a defect introduced by this PR — those gaps existed before — but future work could extend this coverage to other SDKs using the same shared harness infrastructure.

Verdict

This PR maintains cross-SDK API consistency. The BYOK E2E infrastructure is additive and non-breaking.

Warning

Firewall blocked 1 domain

The following domain was blocked by the firewall during workflow execution:

• awmgmcpg

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "awmgmcpg"

See Network Configuration for more information.

Generated by SDK Consistency Review Agent for #2010 · 75.5 AIC · ⌖ 7.6 AIC · ⊞ 5K · ◷

[SteveSandersonMS]

Excellent - this looks extremely nice.

Very glad to see this detected divergences between the behaviors of different providers. We should look into these as follow-ups.